Extracting Features and Classifying Anomalies using Computer Vision and Machine Learning

Join us to explore the fusion of computer vision and machine learning concepts in MATLAB using a real-world application.  In this session you’ll learn the fundamentals of computer vision and machine learning to develop a classifier capable of discriminating normal tissue from cancerous tissue.  To do so, we’ll be working with data captured during a surgical procedure using an endoscopic near-infrared fluorescence imaging system.  This system simultaneously captures and displays videos in both the visible white-light and near-infrared spectrums. 

Using the visible white-light video, you’ll see how to use computer vision to automatically detect and track features points.  These tracked points are then used to perform image registration to stabilize the video.  This is done by computing a transformation to warp the current frame such that features remain aligned with the initial frame.  With each near-infrared video frame aligned to the same reference frame, the mean fluorescence intensity of a grid of regions can then be quickly computed using distinct block processing of the warped intensity frame.

Using features extracted from these fluorescence intensity time signals, you’ll see how to use machine learning to develop a cancer classifier.  To do so, we’ll use the Classification Learner app to perform common supervised learning tasks such as interactively exploring data, ranking and selecting features, specifying validation schemes, training and optimizing models, and assessing results.  We’ll then export the resulting classification model and use it to create a prediction probability heatmap across the full field of view of test videos not used during training.